Multi-camera arrays have been widely used in stereo vision and panoramic video systems. Calibrating the extrinsic parameters of these cameras properly is the prerequisite for running any complex vision or video algorithm, as these algorithms will often fail if the spatial relationship between the cameras is not accurately defined. State-of-the-art camera calibration methods perform well when there is a large overlapping region in the camera's field of view (FOV). However, in cases where the overlapping region is small or even does not exist, the conventional methods tend to fail due to low resolution of the overlapping region, or failure to find a common pattern in the overlapping region.
Therefore, there is a need for robust calibration methods for cameras having small or even no overlapping field of view.